GA Based PHOG-PCA Feature Weighting for On-RoadVehicle Detection

Vehicle detection is an important issue in driverassistance systems and self-guided vehicles that includes twostages of hypothesis generation and verification. In the firststage, potential vehicles are hypothesized and in the secondstage, all hypothesis are verified. The focus of this work is toclassify vehicle candidate images into vehicle and non-vehicleclasses. We extract Pyramid Histograms of Oriented Gradients(PHOG) features from a traffic image as candidates of featurevectors to detect vehicles. Principle Component Analysis (PCA)is applied to these PHOG feature vectors as a dimensionreduction tool to obtain the PHOG- PCA vectors. Then weemploy real coded chromosome Genetic Algorithm (GA) andlinear Support Vector Machine (SVM) to classify thePHOG-PCA features as well as to improve their performanceand generalization. Our tests show good classification accuracyof more than 96% correct classification on realistic on-roadvehicle images.